Compressing model weights and activations to lower precision, for example 8‑bit or 4‑bit, to reduce memory and speed up inference.
Compressing model weights and activations to lower precision, for example 8‑bit or 4‑bit, to reduce memory and speed up inference.
